Skip to content

vllm.models.hy_v4_flash.nvidia.attention

Per-layer MLA projections with Flash's SWA/IndexPool selection.